Mitigating Systematic Errors in Roadside Camera-Based V2X Perception

As Cooperative Intelligent Transport Systems (CITS) continue to evolve, roadside infrastructure equipped with cameras, lidars, and radars to detect and report objects through Collective Perception Messages (CPMs) will play a crucial role in enhancing the situational awareness of connected users within the monitored environment. In the longer term, Connected and Autonomous Vehicles (CAVs) are expected to integrate such information with their on-board sensor data to support decision-making processes that may directly impact road users safety. To enable this, infrastructure-generated messages must provide highly accurate information on detected objects, particularly regarding their kinematic properties and associated uncertainties. However, camera sensors are sensitive to environmental factors such as object distance, position within the Field of View (FoV),weather, and lighting conditions. In this work, we propose a method to dynamically estimate and mitigate systematic errorsin camera-based kinematic state estimation by an infrastructure. The correction mechanism relies on an initial offline deviationanalysis performed using a probe vehicle equipped with high-precision instruments that moves through the camera’s FoV.We evaluate the method under diverse weather and lighting conditions and demonstrate a significant reduction in kinematic estimation deviations after correction.